Decision Tree Pruning for Event Prediction Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Service providers face challenges in efficiently identifying and evaluating features to predict events, such as payment transaction downgrades or chargebacks, due to the dynamic nature of these events and the large number of attributes involved.
Innovation Solution
A data mining system is configured to determine relevant data for predicting events by generating and pruning decision trees based on attribute values associated with activities, thereby reducing the complexity of machine learning models and improving prediction accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a service provider uses traditional machine learning models to predict events such as payment transaction downgrades or chargebacks, then the model can process a large number of attributes, but the model complexity and computational resources required for training and usage increase significantly
Solution Approach 1:
The patent extracts and identifies only the most relevant features from a large set of attributes using decision tree analysis. By taking out the critical features that actually predict events like downgrades and chargebacks, the system reduces model complexity while preserving prediction accuracy. The decision tree methodology automatically selects the subset of features that provide the most predictive power, eliminating unnecessary attributes from the machine learning model.
Solution Approach 2:
The patent segments the large set of attributes into hierarchical groups based on decision tree analysis. By organizing features into segments ranked by their predictive importance, the system can process attributes in a structured manner and select only the most relevant segments for modeling. This segmentation approach reduces the overall complexity by dividing the attribute space into manageable, prioritized groups.
2Adaptability or versatility
If a service provider constantly re-trains and re-configures machine learning models to adapt to dynamic event patterns, then the prediction remains relevant to current conditions, but the time and computational resources required increase
Solution Approach 1:
The patent performs preliminary feature identification and selection using decision tree analysis on historical data before deploying the machine learning model. By pre-identifying the stable, predictive features through decision tree methodology, the system creates a robust feature set that remains valid even as event patterns evolve. This preliminary action reduces the need for frequent model re-training, as the core predictive features tend to be more stable over time.
Solution Approach 2:
The patent implements a dynamic feature selection approach where the decision tree analysis can be periodically re-run to identify changes in feature importance. This allows the system to adapt to evolving event patterns by updating the feature set based on current data, rather than re-training the entire model frequently. The dynamic re-analysis of feature importance enables efficient adaptation with reduced computational overhead.
3Loss of information
If a service provider analyzes all attributes associated with activities to identify causes of events, then comprehensive understanding is achieved, but the computational resources and processing time required increase
Solution Approach 1:
The patent extracts the essential predictive features from the complete set of attributes by applying decision tree analysis. This extraction process identifies and isolates the specific attributes that have the strongest relationship with target events, discarding or de-emphasizing attributes that provide minimal predictive value. The result is a streamlined feature set that maintains the essential information needed for accurate prediction while significantly reducing computational requirements.
Solution Approach 2:
The patent applies different levels of analysis depth to different attribute groups based on their predictive importance. By using decision tree methodology, the system automatically assigns different weights and levels of scrutiny to different features - focusing computational resources on the most informative attributes while using simpler analysis for less critical ones. This local quality approach optimizes the balance between information completeness and computational efficiency.
Data Source
AI summary
Methods and systems are presented for mining data in association with predicting occurrences of events. Upon detecting an occurrence of an event associated with a transaction, a data mining system accesses data associated with different transactions, and generates a decision tree for predicting occurrences of the event based on the data. Using a classification specification, the data mining system traverses the decision tree and prunes at least a portion of the decision tree that does not satisfy the classification specification. The data mining system then extracts data relevant to predicting occurrences of the event from the pruned decision tree. The extracted data includes attributes and/or criteria that are relevant to predicting occurrences of the event. Based on the extracted data, one or more actions can be performed to improve the event prediction process and/or reduce the frequency of the occurrences of the event.